Model evaluation#
This page compares simulated assimilation, transpiration, stomatal conductance, leaf temperature, and energy fluxes with measurements. The figures only show the PlantBiophysics results, but comparison with other models is available from the paper and its companion website.
How to read the figures#
In an observed-versus-simulated plot, points close to the 1:1 line indicate agreement. Each panel reports two measures:
NRMSE is the root mean squared error divided by the observed range. Smaller values indicate less error.
EF is modelling efficiency. A value of 1 indicates perfect agreement; 0 is equivalent to always predicting the mean observation, and negative values indicate a poorer prediction than that mean.
The descriptions below identify which quantities are simulated and which are supplied from measurements. This matters when interpreting a model's performance, especially when stomatal conductance is prescribed.
Global evaluation#
The coupled FvCB–Medlyn–Monteith simulation is evaluated against the Tumbarumba gas-exchange measurements from Medlyn, Pepper, and Keith (2015). The paper's Ca > 150 ppm quality filter leaves 536 observations. Each observation is evaluated once; this avoids the duplicated timestamp matches in the historical paper workflow. The grey line is the 1:1 relationship.
Daily evaluation#
This six-observation sequence is for tree 3 on 14 November 2001 in the same Tumbarumba dataset. The line is the mean of 2,000 uncertainty-propagation particles, and the shaded ribbon spans their 2.5th to 97.5th percentiles.
Measured stomatal conductance is forced in this scenario because these spot measurements do not provide the response curve needed to fit a stomatal model. The figure therefore evaluates the photosynthesis and energy-balance response conditional on measured conductance; it is not an independent validation of stomatal conductance.
Schymanski et al. (2017)#
Points are observations and lines are PlantBiophysics simulations for the wind-speed experiment underlying figure 6a of Schymanski and Or (2017). The plotted fluxes are latent heat (λE), sensible heat (H), net radiation (Rn), and the observed energy sum (H + λE).
Measured stomatal conductance, radiative forcing, and chamber conditions are prescribed. This is consequently a targeted evaluation of the energy-balance implementation rather than a fully independent evaluation of the coupled leaf model. The source data come from the Schymanski_leaf-scale_2016 repository.
To reproduce only this panel, run the standalone Schymanski wrapper from the repository root:
bash
julia --project=docs docs/src/models/schymanski.jlThe wrapper calls the same scenario as the numerical regression tests and the same plotting code as the complete documentation build; it does not maintain a second copy of the scientific computation.
Reproducing the figures#
The SVG files are regenerated at the start of every Documenter build. Generation uses only the committed offline test fixtures. To regenerate the figures without building the rest of the documentation, instantiate the docs environment and run:
bash
julia --project=docs -e 'using Pkg; Pkg.instantiate()'
julia --project=docs docs/figures/generate_evaluation_figures.jlThe figures and numerical regression tests share the same scenarios, inputs, parameters, filters, and unit conversions. The tests compare NRMSE, RMSE, bias, normalized bias, and EF with the paper repository's reference results. Their tolerances allow small numerical differences while detecting a meaningful loss of model performance.
The Medlyn fixtures are distributed under CC BY 4.0. Their file identifiers, checksums, extraction history, and the provenance note for the Schymanski data are recorded in test/inputs/evaluation/README.md.
On this page